Informe final* del Proyecto JM033 spatial measures of model … · Information on species...

Informe final* del Proyecto JM033 Improving species distribution models of endangered plants (Orchidaceae, Pinaceae,

Cupressaceae, Taxaceae, Podocarpaceae) in Mexico by utilizing remote sensing data and spatial measures of model uncertainty - in collaboration with IB - UNAM*

Responsable: Dr. Victor Werner Steinmann

Institución: Helmholtz Zemtrum für Umweltforschung GmbH - UFZ

Dirección: Permoserstraße 15, Leipzig, 04318 , ND

Correo electrónico: [email protected]

Teléfono/Fax: +49 (0)931-31-81865 fax: +49 (0)931-31-87358

Fecha de inicio: Noviembre 18, 2013.

Fecha de término: Diciembre 5, 2017.

Principales resultados: Base de datos, cartografía, informe final.

Forma de citar** el informe final y otros resultados:

Cord, A. y D. Gernandt. 2017. Improving species distribution models of endangered plants (Orchidaceae, Pinaceae, Cupressaceae, Taxaceae, Podocarpaceae) in Mexico by utilizing remote sensing data and spatial measures of model uncertainty - in collaboration with IB - UNAM. Helmholtz Zemtrum für Umweltforschung GmbH - UFZ. Informe final SNIB-CONABIO, Proyecto No. JM033. Ciudad de México.

Resumen:

The project aims to develop and refine species distribution models (SDM) of the 987 plant species included in the Norma Oficial Mexicana NOM-059-SEMARNAT-2010 with the perspective to support conservation management. The SDM refinement for these endangered species includes two major aspects, namely (1) the inclusion of remote sensing data besides climatic data and (2) spatial assessments of model uncertainty. For each species, different products of distribution ranges (continuous, categorical) are modeled at 1 km² resolution using different algorithms (e.g. Maxent, GLM, RandomForest). As remote sensing predictors, Terra-MODIS time series (Enhanced Vegetation Index, Land Surface Temperature, Surface reflectance) which have been shown to be useful for vegetation classification are analyzed. The project also involves importance assessments of climate and remote sensing variables to explain species distributions and the extraction of species-specific phenological profiles. Maps of areas where climatically suitable space is not occupied by a species according to remotely sensed distribution ranges are produced to identify probable hot spots of anthropogenic habitat deterioration in the country. The use of current remote sensing data for species distribution modeling thus paves the way towards a national biodiversity monitoring system.

_______________________________________________________________________________________________

* El presente documento no necesariamente contiene los principales resultados del proyecto correspondiente o la

descripción de los mismos. Los proyectos apoyados por la CONABIO así como información adicional sobre ellos,

pueden consultarse en www.conabio.gob.mx

** El usuario tiene la obligación, de conformidad con el artículo 57 de la LFDA, de citar a los autores de obras

individuales, así como a los compiladores. De manera que deberán citarse todos los responsables de los proyectos,

que proveyeron datos, así como a la CONABIO como depositaria, compiladora y proveedora de la información. En

su caso, el usuario deberá obtener del proveedor la información complementaria sobre la autoría específica de los

datos.

mailto:[email protected]

http://www.conabio.gob.mx/

Comisión Nacional para el Conocimiento y Uso de la Biodiversidad

Final report for the project JM033

I. Project title

“Improving species distribution models of endangered plants (Orchidaceae,

Pinaceae, Taxaceae, Podocarpaceae) in Mexico by utilizing remote sensing

data and spatial measures of model uncertainty“

II. General information

Responsible for the project:

Anna F. Cord, PhD: Helmholtz-Centre for Environmental Research – UFZ, Computational

Landscape Ecology, Permoserstraße 15, 04318 Leipzig, Germany

Collaborators:

David S. Gernandt, PhD: Departamento de Botánica, Instituto de Biología, Universidad

Nacional Autónoma de México, México, D.F., 04510, México (responsible for project JM 078)

Gerardo A. Salazar Chávez, PhD: Departamento de Botánica, Instituto de Biología,

Universidad Nacional Autónoma de México, México, D.F., 04510, México

Michael Ewald, MSc Geoecology: freelance collaborator, Schwetzinger Straße 45, 76139

Karlsruhe, Germany

Eckardt Kasch, MSc Geoecology: freelance collaborator, Sophienthal 85,

95466 Weidenberg, Germany

Björn Reineking, PhD: UR EMGR Écosystèmes Montagnards, Irstea, 38402 St-Martin-

d’Hères, France

Leipzig, September 2017

2

III. Summary (Resumen ejecutivo)

The aim of this project – carried out as collaboration between the Helmholtz Centre for

Environmental Research-UFZ (Germany) and the Institute of Biology of the National

Autonomous University of Mexico (UNAM) – was to develop and refine species distribution

models (SDMs) of 227 plant species included in the Norma Oficial Mexicana NOM-059-

SEMARNAT-2010, belonging to the families Orchidaceae (188 species), Pinaceae (30

species), Cupressaceae (7 species), Taxaceae (1 species), and Podocarpaceae (1 species). The

SDM refinement for these endangered species included two major aspects, namely (1) the

inclusion of remote sensing variables (Terra-MODIS Enhanced Vegetation Index and Land

Surface Temperature) in addition to climatic predictors and (2) the spatially-explicit assessment

of model uncertainty. For each species, different products of distribution ranges (continuous

and categorical) were modeled at 1 km² resolution using different algorithms (Maxent, GLM,

and Random Forest). The project also involved importance assessments of climate and remote

sensing variables to explain species distributions and the extraction of species-specific

phenological profiles. Maps of areas where climatically suitable space is not occupied by a

species according to remotely sensed distribution ranges were produced to identify probable

hot spots of anthropogenic habitat deterioration for the modeled species.

Only for a subset of the 227 species initially considered in the project enough unique

species records could be compiled from existing databases and herbarium collections (done in

the context of project JM078) to allow reasonable SDMs. In total, predictions could therefore

be made based on climate data for 41 species and based on remote sensing data for 31 species,

respectively. Because of the limited number of species that could be modeled based on remote

sensing data, not enough results were available to further investigate the potential of remote

sensing data in SDMs based on species characteristics in a conceptual framework, which was

initially planned. However, once additional records will be available in the future, the way the

projected was implemented in the R programming language would allow re-running models for

the additional species. Remote sensing data showed great potential for reducing the typical

overprediction of climate-based SDMs. Even though also predictions based only on climate

data were produced in the project, we therefore highly recommend using the predictions for

habitat availability (which are based on remote sensing data but also consider limitations to

potential species distribution ranges based on climate) for further analyses.

3

IV. Table of contents (Indice)

I. Project title .............................................................................................................................. 1

II. General information ............................................................................................................... 1

III. Summary (Resumen ejecutivo) ............................................................................................ 2

IV. Table of contents (Indice) .................................................................................................... 3

V. Introduction (Introducción) ................................................................................................... 4

VI. Project development (Desarrollo del Proyecto) ................................................................... 5

1. Quality of information (Calidad de Información) .............................................................. 5

2. Criteria for the selection of the reference area per species (Criterio de Selección de la región

de referencia por especie) ....................................................................................................... 9

3. Variables used in the models (Variables utilizadas en la modelación) .............................. 9

4. Methods for modeling (Método de modelación) .............................................................. 12

5. Parameters used for modeling (Parámetros utilizados en la modelación) ........................ 14

6. Model evaluation (Evaluación del modelo) ...................................................................... 15

7. Conclusions and recommendations (Conclusiones y recomendaciones) ......................... 19

8. References ........................................................................................................................ 21

9. Appendix .......................................................................................................................... 23

1. Documentation of delivered materials (Documentación del material entregado) ........ 23

2. Quality of information (Calidad de información) ......................................................... 35

3. Other information .......................................................................................................... 36

4

V. Introduction (Introducción)

Particular attention has been paid to climatic parameters and the potential impacts of changing

climate on species distributions for a wide range of taxa, study areas and spatial scales (e.g.

Araújo et al., 2005). However, despite its importance, the impact of anthropogenic land use/land

cover (LULC) change is often not considered in species distribution modeling. We know that

climate governs species distributions at coarse (global, continental) scales whereas LULC is a

main aspect for presence of species at finer (regional to local) scales (Luoto et al., 2007).

Therefore, a hierarchical framework for the interaction of climate and LULC has been proposed

and successfully applied in species distribution modeling (Pearson et al., 2004; Thuiller et al.,

2004). Even though the incorporation of LULC data can allow identifying regions with suitable

climate but unsuitable land cover, LULC maps are often not (thematically) detailed enough to

improve predictions of species distributions (Bradley & Fleishman, 2008). In line with recent

studies (e.g. Buermann et al., 2008; Saatchi et al., 2008; Tuanmu et al., 2010), we therefore

used satellite imagery instead of LULC data as predictors in this project to model plant species

distributions in Mexico. Especially the analysis of remote sensing time series appears promising

to describe vegetation and habitat characteristics.

Uncertainty in the modeled probabilities of occurrence (arising from several sources such

as the quality of occurrence and environmental data and the choice of model algorithm) can be

substantial and spatially structured, and uncertainty maps are an important tool for

communicating the extent and spatial patterns of uncertainty (Elith et al. 2002). Building on

previous case studies (Dormann et al. 2008, Buisson et al. 2010), we therefore also provide

information about the reliability of the species distribution predictions. To account for

prediction uncertainty due to the choice of model algorithm, several algorithms (Maxent,

Generalized Linear Models and Random Forest) were employed and combined in so-called

‘ensemble models’.

The study species considered in this project (Orchidaceae: 188 species; Pinaceae: 30

species; Cupressaceae: 7 species: Taxaceae: 1 species; Podocarpaceae: 1 species, Appendix A1)

are endangered or at risk of extinction according to the official NOM-059-SEMARNAT-2010

standard. Especially for those species threatened by habitat loss, maps of potential distributions

based solely on climate data are insufficient to support sustainable conservation management

and decision making. Therefore, reliable and detailed estimates of their current distributions

(based for example on remote sensing data) and spatial indicators of model uncertainty are

required to target conservation efforts.

The main outcomes and novel technical aspects of this project were: (1) the use of a

hierarchical modeling scheme, i.e. the implementation of species distribution models separately

for climatic and remote sensing data and the subsequent combination of model predictions. (2)

The species-specific selection of predictors based on predictor collinearity and explanatory

power together with the analysis of variable importance of different products of remote sensing

data. (3) Spatially explicit uncertainty assessments for predicted probabilities of occurrence for

each modeled species.

5

VI. Project development (Desarrollo del Proyecto) 1. Quality of information (Calidad de Información)

Species records. Species records used in this project were collected within the project JM078

(lead by Dr. David S. Gernandt, UNAM). For this purpose, available presence records of the

227 conifer and orchid species considered in the NOM-059-SEMARNAT-2010 (Orchidaceae:

188 species; Pinaceae: 30 species; Cupressaceae: 7 species: Taxaceae: 1 species;

Podocarpaceae: 1 species) were compiled from existing databases and herbarium collections.

The existing databases considered here were collected by Dr. David Sebastian Gernandt and

Dr. Gerardo Adolfo Salazar Chávez, including the running project "Digitalización y

Sistematización de las Colecciones Biológicas Nacionales" (KE02; Head: Dr. Víctor Manuel

Sánchez-Cordero Dávila). The specimens in the resulting database of this project stem mainly

from the following herbaria: the Herbario Nacional of Mexico (MEXU), the Herbario AMO

(AMO), the California Academy of Sciences (CAS), the University of California, Berkeley

(UC) and the New York Botanical Garden (NY). To map the distribution of non-endemic

species of Mexico, also a limited number of records from GBIF (http://data.gbif.org) for species

with distributions in Canada, the United States, Belize, Guatemala, Honduras and El Salvador

were considered. The quality assessment of species records included cross referencing against

existing catalogs, the analysis of typographic and taxonomic problems and the verification of

their plausibility in geographical space. For those records for which no coordinates were

available, georeferencing was done using locality information from herbariums labels, Google

Earth and literature review. Information on species localities from SNIB (Sistema Nacional de

Información Sobre Biodiversidad) was not included in the project as initially planned because

of many taxonomic inconsistencies highlighted by the taxonomists involved. More detailed

information on the species database can be found in the report of project JM078.

In order to avoid pseudo-replication in the species distribution models, duplicate

records, i.e. records within the same pixel of the environmental data (WorldClim climate data,

MODIS remote sensing data) were removed. Further, species records for which no

environmental data were available were not considered. Based on the recommendations of

Hijmans & Elith (2015), a minimum sample size of 20 presence localities for model building

was accepted. Species with less than 20 records were therefore not modeled (see Appendix A1).

This resulted in 3,162 records used for the climate-based models and 1,274 records used for the

remote sensing-based models, respectively (Table 1, Figure 1). A full overview of records used

in this project (in DarwinCore format) analogue to the results of project JM078 was submitted

together with this report.

Table 1. Overview of species records used for modeling.

Family

Number of records used in

climate-based models (number

of species)

Number of records used in

remote sensing-based models

(number of species)

Orchidaceae 653 (17) 640 (17)

Pinaceae 2,050 (17) 393 (10)

Cupressaceae 316 (5) 130 (2)

Taxaceae 51 (1) 51 (1)

Podocarpaceae 52 (1) 52 (1)

Total 3,122 (41) 1,266 (31)

http://data.gbif.org/

6

Figure 1. Location of presence records used in the species distribution models for the different

plant families considered. (a) For climate-based models, (b) for remote sensing-based models.

Bioclimatic profiles. Bioclimatic profiles of the modeled species were compiled based on

WorldClim climate data (Hijmans et al., 2005; see Fig. 2 for an example for Pinus nelsonii and

Podocarpus matudae). The corresponding figures for the remaining species are shown in

Appendix A2. Summarizing tables considering all WorldClim bioclimatic variables can be

found in Appendix A3.

7

Figure 2. Bioclimatic profiles of (a) Pinus nelsonii and (b) Podocarpus matudae based on

WorldClim climate data (Hijmans et al., 2005; current conditions).

Profiles of land surface phenology. Phenological profiles of the modeled species were derived

from time series data of the Terra-MODIS Enhanced Vegetation Index (EVI) from the years

2001 to 2009 (for further details: see Section 3, Variables used in the models). For each species,

EVI values were extracted from the raster maps at the respective presence locations. Extracted

values were averaged for each date over all locations (Figure 3a) as well as stratified by

biogeographic region (Figure 3b; according to CONABIO, 1997) and vegetation type (Figure

(a)

(b)

8

3c; Rzedowski, 1978; Rzedowski, 1990). Both mean values and standard deviation were plotted

against time (see Appendix A4 for plots of the remaining species).

Figure 3. Phenological profiles of Pinus nelsonii. (a) Summarized over all locations, (b)

stratified by biogeographic region, and (c) stratified by vegetation type.

(a)

(b)

(c)

9

2. Criteria for the selection of the reference area per species (Criterio

de Selección de la región de referencia por especie)

Because of the large number of species considered in this project, we had - in consultation with

CONABIO - to refer to a fixed geographical extent. While the study area for modelling

available habitat based on remote sensing data was confined to Mexico, the area for modelling

potential species distributions based on climate data was set to a rectangular area of 10 degrees

(WGS84, in all four directions) beyond the extent of Mexico (according to the GADM database

of Global Administrative Areas, GADM 2012).

3. Variables used in the models (Variables utilizadas en la

modelación)

Climate data. Climatic variables were obtained at 30 arc seconds (~ 0.0083 degree) spatial

resolution from the WorldClim data base (Current conditions; Hijmans et al., 2005). WorldClim

parameters are derived from long-term time series (1950-2000) of a global network of climate

stations, express spatial variations in annual means, seasonality, and extreme or limiting

climatic factors, and represent biologically meaningful variables for characterizing species

distributions. For the species distribution models, bioclimatic layers were resampled from their

native resolution and gridded to pixel location, cell size and extent of the remote sensing data

to maintain spatial consistency. This was done with R 3.0.1 (R Core Team 2013) using the sp

package version 1.0 (Pebesma & Bivand, 2005).

Remote sensing data. The remote sensing parameters used in this project were selected to

effectively describe vegetation dynamics and land surface phenology (Colditz et al., 2009; Cord

& Rödder, 2011). Vegetation indices are an integrated measure of vegetation canopy greenness,

a composite property of leaf area, chlorophyll and canopy structure as important dimensions of

habitat characteristics. We used the Enhanced Vegetation Index (EVI) that is superior to the

Normalized Difference Vegetation Index (NDVI) due to its improved sensitivity in high biomass

regions and a reduction in atmospheric influences (Huete et al., 2002). In addition, Land Surface

Temperature (LST) as one of the key parameters in the physics of land surface processes, e.g.

surface-atmosphere interactions and energy fluxes, was analyzed. LST provides different

information compared to temperature measures derived from interpolated climate station data.

Remote sensing time series from 2001 to 2009 at 1 km² spatial resolution based on the

Terra-MODIS 16-day standard products MOD13A2 and MOD11A2 were compiled for

Mexico. For this purpose, the MODIS-specific pixel-level Quality Assurance Science Data Sets

(QA-SDS) were analyzed using the TiSeG software package (Colditz et al., 2008) to exclude

low-quality data, e.g. due to cloud cover or atmospheric contamination. With a critical

weighting between data quality and the necessary quantity for meaningful interpolation, high-

quality data were used as vertices for pixel-level linear temporal interpolation. In addition, an

adaptive Savitzky-Golay filter as implemented in the TIMESAT 3.0 software (Jönsson &

Eklundh, 2004; Eklundh & Jönsson, 2009) was applied to account for high-frequency

fluctuations and negatively-biased noise.

From the time series data, 18 annual phenological metrics including (1) Time-related

metrics: Start of season (date_SOS), mid of season (date_maximum), end of season (date_EOS),

dormancy (date_dormancy), length of season (length_season), (2) Net primary productivity

(NPP)-related metrics: Vegetation index value at SOS (value_SOS), value at EOS (value_EOS),

maximum value (maximum), minimum value (minimum), annual range (range), accumulated

integral during vegetation period (integral), annual mean (mean), annual median (median) and

(3) Seasonality-related metrics: rate of green-up (rate_greenup), rate of senescence

10

(rate_senescence), shape of phenology curve (skewness), standard deviation

(standard_deviation) and coefficient of variation (CoV). For temporal metrics referring to

certain stages within the phenological cycle, the number of the corresponding composite

(between 1 and 23 in accordance with the 16-day composite period of the MODIS products)

was assigned. In addition, annual statistical metrics (minimum, mean, median, maximum,

range, standard deviation, and coefficient of variation) were computed for the Land Surface

Temperature (LST) time series. All metrics were averaged over the nine year period to account

for inter-annual variation of vegetation seasonality or single-year anomalies before they were

used as predictors in the species distribution models.

Correlation analysis. To avoid model over-fitting and to exclude redundant data,

environmental predictors (both climate and remote sensing) were carefully selected. For both

data sets, species-specific pair-wise Spearman’s rank correlation coefficients were estimated.

Out of each pair of highly-correlated and hence redundant environmental predictors

(Spearman’s rank correlation coefficient |r| > 0.7) the variable with the higher explanatory

power for each study species according to a Generalized Linear Model (GLM) based on

presence and pseudo-absence locations was retained (select07 method in Dormann et al., 2013

which selects variables based on removing correlations > 0.7, retaining those variables more

important). However, because models still tended to be over-fitted in trial runs with these

reduced predictor set for species with more than 50 occurrence records, we decided to exclude

the time-related variables (EVI_date_SOS, EVI_date_EOS, EVI_date_maximum, EVI_date

dormancy and EVI_length_season) due to their low mean influence (relative variable

importance in most cases below 0.1) in the trial runs. Furthermore, because of its low ecological

interpretability the variable EVI_skewness was excluded. Collinearity among the remaining 19

remote sensing variables (Table 2) was analyzed using the select07 method in Dormann et al.

(2013) as described above. An overview of how often the variables were selected is given in

Appendix A5 (for climate) and A6 (for remote sensing).

11

Table 2. Final set of variables used in the species distribution models.

Short name (Abbreviation) Description C

lim

ate

BIO1 Annual Mean Temperature

BIO2 Mean Diurnal Range (Mean of monthly (max - min temp))

BIO3 Isothermality (BIO2/BIO7) (* 100)

BIO4 Temperature Seasonality (standard deviation *100)

BIO5 Max Temperature of Warmest Month

BIO6 Min Temperature of Coldest Month

BIO7 Temperature Annual Range (BIO5-BIO6)

BIO8 Mean Temperature of Wettest Quarter

BIO9 Mean Temperature of Driest Quarter

BIO10 Mean Temperature of Warmest Quarter

BIO11 Mean Temperature of Coldest Quarter

BIO12 Annual Precipitation

BIO13 Precipitation of Wettest Month

BIO14 Precipitation of Driest Month

BIO15 Precipitation Seasonality (Coefficient of Variation)

BIO16 Precipitation of Wettest Quarter

BIO17 Precipitation of Driest Quarter

BIO18 Precipitation of Warmest Quarter

BIO19 Precipitation of Coldest Quarter

Rem

ote

sen

sin

g

LST_standard_deviation (LSTSTD) Standard deviation of annual mean land surface

temperature

LST_range (LSTRANGE) Range of annual land surface temperature

LST_minimum (LSTMIN) Minimum annual land surface temperature

LST_median (LSTMEDIAN) Median annual land surface temperature

LST_mean (LSTMEAN) Annual mean land surface temperature

LST_maximum (LSTMAX) Minimum annual land surface temperature

LST_CoV (LSTCOV) Coefficient of Variation of land surface temperature

EVI_value_SOS (EVIVSOS) Mean Enhanced Vegetation Index at beginning of season

EVI_value_EOS (EVIVEOS) Mean Enhanced Vegetation Index at end of season

EVI_standard_deviation (EVISTD) Standard deviation of the mean annual value of the

Enhanced Vegetation Index

EVI_rate_senescence (EVISENES) Rate of senescence

EVI_rate_greenup

(EVIGREENUP) Rate of greenup

EVI_range (EVIRANGE) Annual range of the Enhanced Vegetation Index

EVI_minimum (EVIMIN) Minimum annual value of the Enhanced Vegetation Index

EVI_median (EVIMEDIAN) Median annual value of the Enhanced Vegetation Index

EVI_mean (EVIMEAN) Mean annual value of the Enhanced Vegetation Index

EVI_maximum (EVIMAX) Maximum annual value of the Enhanced Vegetation Index

EVI_integral (EVIINTEG) Sum of Enhanced Vegetation Index values between start

and end of season

EVI_CoV (EVICOV) Coefficient of Variation of the Enhanced Vegetation Index

12

4. Methods for modeling (Método de modelación)

Generation of pseudo-absence data. Pseudo-absence data were generated following the

target-group background approach (Phillips et al., 2009). According to this assumption, the

influence of spatially biased samples (e.g. towards roads and protected areas, which is typical

for biological collections) can be reduced by comparing the occurrences with background points

reflecting the same spatial bias. The underlying idea is that a model based on presence and

background data with the same bias will not focus on the sample selection bias, but on any

differentiation between the distribution of the occurrences and that of the background. Because

the species analyzed in this study are from different families, all species records belonging to

the kingdom Plantae within the study area extent were extracted from the GBIF database

(http://data.gbif.org) (in total 1,990,527 records). From this dataset, two randomized subsets of

10,000 unique locations were selected as pseudo-absence data sets for the potential species

distributions (based on climate data) and the extent of available habitat (based on remote

sensing data) (Figure 4 a and b).

Figure 4. Location of pseudo-absence records. a) For climate-based models, b) For remote

sensing-based models.

Model algorithms and ensemble modeling. Since the choice of model algorithm is a major

component of prediction uncertainty in species distribution modeling (Dormann et al., 2008;

http://data.gbif.org/

13

Buisson et al., 2010), several algorithms were employed in order to estimate and account for

prediction uncertainty. The modelling procedure was conducted for both, potential distribution

and available habitat, using the same methodology. If not mentioned otherwise, model

adjustments were identical for both. All modelling procedures were performed using R version

3.1.2 (R Core Team 2014). The final models were run on a Linux-based cluster with 1,024

computing cores and 5TB RAM with a CentOS 6.5 as operating system.

We used three modeling methods that are representative for different classes of model

algorithms, namely regression-based methods, machine-learning methods and presence-only

methods. Generalized Linear Models (GLM), Random Forest (RF) and Maxent (Maximum

Entropy) version 3.3.3e were applied as implemented in the biomod2 package version 3.1

(Thuiller et al., 2014). Presences and pseudo-absences were weighted equally and were used as

response variables. Model runs were repeated 199 times for each species and model algorithm,

randomly selecting 70% of the presences and pseudo-absences for model calibration and 30%

for model testing in each model run. Models were scaled between 0 and 1,000 to ensure

comparability of model predictions derived from the different model algorithms. To generate

ensemble (“consensus”) models from the results of these three algorithms, the ROC (Relative

or Receiver Operating Characteristic) was used as evaluation metric (see the documentation of

the biomod2 package for further information). For species with more than 100 occurrences (see

below), additional model predictions were made using Generalized Additive Models (GAM)

and Boosted Regression Trees (BRT). These additional models were done with 10 repetitions.

The specific parameters used for each modeling algorithm are described in further detail in

section 5 ‘Parameters used for modeling’.

Conversion to binary maps and combination of climate and remote sensing-based model

predictions. From the resulting consensus maps (i.e., based on GLM, RF and Maxent; see

above), binary species-specific distribution maps were created by applying three different

threshold values that give different weight to omission and commission errors: The ‘minimum

training presence’, ‘10 percentile training presence’, and ‘maximum training sensitivity plus

specificity’ threshold. The ‘minimum training presence threshold’ (representing the lowest

value observed in the continuous prediction map at a presence location for a specific species)

aims at minimizing omission errors at the expense of a greater fractional predicted area per

species and may lead to model overprediction (Jarnevich & Reynolds, 2011). The ‘10 percentile

training presence threshold’ provides more conservative models by predicting the 10% most

extreme presence observations as absent. Finally, the ‘maximum sensitivity and specificity

threshold’ gives equal weight to both omission and commission errors (sensitivity = true

positive rate and specificity = true negative rate of the predictions) and aims at maximizing

both. It was identified as among the best approaches by Liu et al. (2005). Sensitivity and

specificity values were calculated using the PresenceAbsence package version 1.1.9 (Freeman

& Moisen, 2008).

Climatic and remote sensing-based species distribution models were finally integrated

in a hierarchical modeling framework where first the potential distribution range (based on the

‘10 percentile training presence’ threshold) was modeled based on climate data and then

remotely sensed habitat availability within this climatic range was integrated in a hierarchical

model design. These maps of current habitat availability were only produced for the Mexican

national territory. We finally used binary maps based on the ‘10 percentile training presence’

threshold to calculate maps of habitat loss by subtracting available habitat from the potential

distributions.

14

5. Parameters used for modeling (Parámetros utilizados en la

modelación)

Maxent. Maxent models were run using auto features, but excluding threshold and hinge

features, with the number of iterations limited to 500. To optimize model performance, the

regularization parameter β (to be applied to all linear, quadratic and product features; Phillips,

2006) was determined in trial runs for species with more than 80 occurrences for climate-based

models (Pseudotsuga menziesii glauca, Abies concolor, Cupressus lusitanica, Calocedrus

decurrens, Pinus attenuata, Pinus jeffreyi, Pinus coulteri, Rhynchostele cervantesii, Pinus

muricata, Prosthechea vitellina) and for species with more than 50 occurrences for remote

sensing-based models (Cupressus lusitanica, Pinus pinceana, Podocarpus matudae,

Prosthechea vitellina, Pseudotsuga menziesii glauca, Rhynchostele cervantesii, Taxus

globosa). For this purpose, we compared model performance using regularization multipliers

of 0.001, 0.002, 0.005, 0.01, 0.02, 0.05, 0.25, 0.5, 0.75, 1, 2, 5 and 10 separately for climate and

remote sensing variables. For the final models, the regularization parameter with the highest

mean AUC value in these trial runs was chosen (climate: β=0.02, remote sensing: β=0.002).

Generalized Linear Models. Generalized Linear Models (GLM) were fitted using a binomial

link function, including linear and quadratic terms as well as linear interaction terms of the

environmental predictors. Interactions considered in the models were chosen based on their

ecological meaningfulness which implied that interactions among temperature variables and

interactions among precipitation variables were not allowed. A stepwise backward selection

procedure starting with the full model was applied to select final models (MASS package,

Venables & Ripley 2002). In each step, the model was simplified and the Bayesian Information

Criterion (BIC) of the simplified model was compared to the BIC of the previous model. This

procedure was repeated until the BIC reached its minimum. Simplification was done by

dropping the variable which leads to a maximum decrease of BIC compared to the starting

model.

Random Forest. As for Maxent, algorithmic settings for RF models were determined in trial

runs for species more than 80 occurrences for climate-based models and for species with more

than 50 occurrences for remote sensing-based models. For this purpose, the number of trees

was tested between 500 and 3,000 in six steps and the number of tries was tested between 1 and

5 in five steps (again separately for climate and remote sensing variables). For each of these 30

different settings, the mean R² of the ‘out-of-bag error estimate’ (R²oob) over all test species

was calculated. The oob is the average misclassification over all trees tested by using the out-

of-bag examples (out-of-bag examples are the data that are left after bootstrapping the training

data; Livingston, 2005). The settings with the highest mean R²oob were chosen for the final

models (Leutner et al. 2012). Based on these results, the following settings were applied:

Climate data: 2,500 trees, 2 tries; Remote sensing data: 500 trees, 1 try.

Generalized additive models (GAM) and boosted regression trees (BRT). As for the other

algorithms, algorithmic settings for BRT models were determined in trial runs for species more

than 80 occurrences for climate-based models and for species with more than 50 occurrences

for remote sensing-based models. For this purpose, the number of trees was tested between

1,500 and 2,500 in three steps. The tree complexity parameter was tested between 3 and 5 in

three steps. The learning rate was tested between 0.005 and 0.008 in four steps. The bag fraction

was tested between 0.5 and 0.8 in four steps. For each of these 144 different settings, the mean

of deviance over all test species was calculated. This was done separately for climate and remote

sensing variable. Based on these results, the following settings were applied: climate: number

15

of trees = 2,500, tree complexity = 5, learning rate = 0.007 and bag fraction = 0.8; remote

sensing: number of trees = 1,500, tree complexity = 4, learning rate = 0.005 and bag fraction =

0.8. All other parameters of BRT models were set to default. The smoothing parameter k in the

GAM models was set to 30. To allow for a comparison among the different model algorithms,

predictions were made for the species Abies concolor, Calocedrus decurrens, Cupressus

lusitanica, Pinus attenuata and Pseudotsuga menziesii glauca using the GLM, RF, Maxent,

GAM and BRT algorithms with 10 repetitions and the model settings described above. Results

showed no significant differences to models run with only GLM, RF and Maxent.

6. Model evaluation (Evaluación del modelo)

Variable importance. Variable importance for GLM, RF, GAM and BRT models was

calculated using the implemented algorithm of the package biomod2 (function

‘variables_importance()’). This method is based on a comparison of the model prediction

derived from the original dataset and predictions derived from permuted datasets. These

permuted datasets were created by randomizing one environmental variable for each data set.

The predicted values of permuted datasets and the original dataset for presence/pseudo-absence

locations were compared by calculating the Pearson correlation coefficient. This procedure was

repeated three times for each variable. The smaller the correlation coefficient was the higher

was the independent influence of the permuted variable. Values of variable importance were

finally calculated by subtracting the correlation coefficient from 1. For the Maxent models,

variable importance was additionally expressed by the regularized model gain for each variable.

For all three algorithms (GLM, Maxent and RF), BIO4 (Temperature seasonality), BIO14

(Precipitation of Driest Month) and BIO15 (Precipitation Seasonality) showed the highest

variable importance among the climatic predictors considered. On average, 3.19 ± 0.65 (mean

± standard deviation) remote sensing variables were selected based on the collinearity and

variable importance analysis (select07 method) per species. These belonged in 11.6% to EVI-

based and in 25.8% to LST-based variables. While there was a high variability regarding

variable importance among the different study species, minimum LST turned out to have the

highest average variable importance for all three algorithms.

Model performance based on AUC, partial AUC and model deviance. We quantified model

performance by calculating AUC values, partial AUC values and model deviance. The AUC

value is the area under the Receiver Operating Characteristic curve (ROC) and describes the

ability of the model to discriminate between 0 and 1, or absence and presence respectively

(Peterson et al., 2008). The ROC is created by plotting the fraction of true presences out of

observed presences (sensitivity) against the fraction of false presences out of observed absences

(1-specificity), for different threshold levels. AUC values were calculated using the package

PresenceAbsence version 1.1.9 (Freeman, 2008). The partial AUC values gives the ratio of the

AUC of the model to the AUC of a random model in a defined range of sensitivity or specificity,

were the model is supposed to make predictions. We used the range of sensitivity between 0.2

and 1, following Peterson et al. (2008) who recommend restricting AUC calculations “to the

domain within which omission error is sufficiently low as to meet user-defined requirements of

predictive ability”. The partial AUC values were calculated by using the package pROC version

1.7.3 (Robin et al., 2011). The model deviance is a log-likelihood ratio statistics that compares

the saturated model with the proposed model. The model deviance was calculated as: D=-

2*(yi*log(ui))+((1-yi)*log(1-ui))n, where yi is the binary observed presence at a locations, ui is

the predicted probability at a location and n is the number of all presences and absences. In

addition, we used presence-absence records from the Mexican National Forest Inventory

(INFyS, Inventario Nacional Forestal y de Suelos) for model validation for those species that

16

were included in the INFyS database. For this purpose, AUC and model deviance were

calculated by using the INFyS presence-absence data as true presence-absence data. All

climate-based models showed very good performance measures based on AUC (i.e. > 0.9) and

partial AUC (i.e. > 0.7). For those species for which an additional model evaluation based on

INFyS data was possible, the obtained (partial)AUC scores were much lower, especially for

models using GLMs. Model performance of the remote sensing-based models was lower with

only 54% of the models having AUC scores of 0.9 or higher. 94% of the models, however,

showed AUC scores above 0.7 and partial AUC scores above 0.5. Again, model deviance was

highest for GLMs.

Exemplary model results. As two examples of a ‘good’ and a ‘bad’ model, the modeled habitat

availability of Pinus nelsonii and Podocarpus matudae are shown in Figure 5. The known

distribution of P. nelsonii (according to Farjon and Filer, 2013) is limited to Coahuila, Nuevo

León, San Luis Potosí and Tamaulipas, in foothills or lower slopes of the Sierra Madre Oriental

– which is well-captured by the model prediction. The species is restricted to sites on rocky

limestone with shallow soils (Farjon and Filer, 2013). Apparently, while climate-based SDMs

tended to overestimate distribution ranges for this species (data not shown), the inclusion of

remote sensing data allowed identifying those sites with suitable habitat conditions. In addition

to the distribution in Coahuila, Nuevo León, San Luis Potosí and Tamaulipas, the model for

habitat availability also predicted potential habitat in Hidalgo and Puebla, though with lower

suitability scores. The model of habitat suitability for P. matudae, however, overestimated

habitat availability, in particular on the Yucatán peninsula. The species occurs from Honduras

to Tamaulipas and Jalisco in Mexico and is mostly found in mixed pine forest, pine-oak forest,

montane rain forest and evergreen cloud forest at altitudes between 600 and 2,600m (Farjon

and Filer, 2013). Such habitat types do not exist on the Yucatán peninsula. However, even

though only one record from Jalisco was included in this project, the model correctly predicted

habitat availability in the Sierra Madre del Sur – Farjon and Filer (2013) list herbarium records

of P. matudae from this region.

17

Figure 5. Modeled habitat availability of (a) Pinus nelsonii and (b) Podocarpus matudae. The

inset map in (a) shows the main distribution range of P. nelsonii on karst limestone outcrops

in the Sierra Madre Oriental.

Spatial assessment of model uncertainty. Maps of spatially explicit model uncertainty were

derived by calculating the difference between the 97.5-percentile and the 2.5-percentile of cell

values derived from the continuous prediction maps, based on 199 model repetition runs. For

species with 40 or more independent presence locations, uncertainty maps were calculated

using the predictions of the Maxent, GLM and RF models. For species with less than 40

independent presence locations, only the Maxent predictions were considered. Results showed

that model uncertainty was generally higher in areas with high predicted suitability and lowest

in those region with the lowest suitability (i.e. that all model algorithms and repetitions

consistently predicted low suitability). In the case of Podocarpus matudae (Figure 6), the

uncertainty map revealed that the overestimated species distribution on the Yucatán peninsula

was modeled with comparatively high uncertainties. Uncertainty maps therefore can effectively

help refining species distribution predictions.

18

For three species with a wide spatial distribution all over Mexico (Cupressus lusitanica, Pinus

strobiformis, Pseudotsuga menziesii glauca), a spatially stratified resampling scheme was

implemented. For each of those species, we performed a k-means clustering (Hartigang &

Wong, 1979) using a maximum number of 10 iterations on the presence/pseudo-absence data

to derive 199 center points. Using each of the center points, the original dataset was divided

into 199 calibration datasets and one validation dataset. During each iteration, 30% of the

presence/pseudo-absence points with the smallest distance to the center points were chosen as

validation data and the remaining 70 % of the points as calibration data. Our aim here was to

test for the relative contributions of different components of model uncertainty (occurrence

data, environmental data, and model algorithm) using generalized linear models (according to

Buisson et al., 2010). For this purpose, the mean predicted probability of each replicate model

run was used as response and the used model type, data type (climatic or remote sensing) and

replicate number as explanatory variables (glm(formula = x ~ data + occurence + modeltype,

family = gaussian). For all species, we found significant effects (p < 0.001) of the environmental

data (i.e. climate or remote sensing) and the model algorithm (Maxent, GLM, RF) used, but not

for the occurrence data (i.e. random vs. spatially stratified resampling, p > 0.05).

Figure 6. Uncertainty map for modeled habitat availability of Podocarpus matudae.

19

7. Conclusions and recommendations (Conclusiones y

recomendaciones)

The project aims, i.e. the development and refinement of species distribution models by (1) the

inclusion of remote sensing variables (Terra-MODIS Enhanced Vegetation Index and Land

Surface Temperature) in addition to climatic predictors and (2) the spatially-explicit assessment

of model uncertainty could be achieved. The main limitation was the limited number of verified

herbarium records available as training data – a typical phenomenon for very rare species as in

the NOM – which did allow model building only for a fraction of the species considered. More

efforts therefore have to be undertaken to compile reliable information on species presence that

is readily accessible online.

Remote sensing data showed great potential for reducing the typical overprediction of

climate-based SDMs. Even though also predictions based only on climate data were produced

in the project, we therefore highly recommend using the predictions for habitat availability

(which are based on remote sensing data but also include potential distributions based on

climate) for further analyses. SDMs based on remote sensing data only, however, also in most

cases overestimated distribution ranges. The combination of climate and remote sensing data

therefore appears to be the most promising approach for modeling species distributions.

However, species records were mostly collected during a different time period than the

remote sensing data. More specifically, 314 records were collected between 2001 and 2009

(which is exactly the time period that the remote sensing data cover), 57 records were collected

after 2009, and 2,294 records were collected before 2001. Information concerning the collection

date was missing for 497 records. In summary, the majority of records were hence collected

before satellite data was available, with a mean collection year of 1968. How big the impact of

this temporal mismatch on the applicability of the model predictions is largely depends on the

characteristic of the collection sites. If specimen were collected in a site where almost no change

in vegetation cover has happened over the last few decades and where the species is still present

today, effects on the resultant prediction maps will be minimal. As many collectors of field data

tend to sample undisturbed and well-established populations, this will hold true at least for a

part of the records. On the other hand, in case land use/land cover change has happened in the

collection site (making the site no longer suitable for the species of concern) between the

collection date of the records and the acquisition date of the remote sensing data, this record

will be unsuitable for model training and should be left out. Generally, SDMs and the

algorithms used in this project can deal with some outliers or wrong records in the training data;

however, models will be significantly biased, if a large fraction of the records is affected by

this. Apparently, it is not possible to quantify the overall effects of the temporal mismatch of

species records and remote sensing data based on the data at hand. A rigorous approach would

require a lot of additional field work in order to collect recent data in those collection sites that

could be affected by land use/land cover change. We therefore recommend taking this into

account when interpreting remote sensing-based model results.

Further, the compiled maps of habitat loss are an indicator of anthropogenic impacts but

should not be overinterpreted and directly be used for decision-making. Other factors not

captured by climate but by remote sensing data (e.g. natural disturbances) may lead to areas

being identified as ‘lost habitat’ independent of any human impact. Obviously, the map

indicating total habitat loss (a summary of the species-specific habitat loss maps) is biased in

the way that it considers only a small number of species and therefore is not representative for

the entire country.

For some species, the different model algorithms used showed large differences

regarding model performance (AUC, partial AUC and model deviance). We therefore

recommend using ensemble models based on different algorithms for species distribution

20

modeling. The high variable importance of LST-derived parameters again supports the

importance of considering additional predictors beyond the ‘classical’ vegetation indices in

SDMs. Further research is needed to fully explore the relevance of the broad array of remotely

sensed parameters available for use in SDMs.

In the future, remotely sensed time series will be operationally available at much higher

spatial resolutions (e.g. from the ESA Sentinel satellites) and provide a unique opportunity for

assessing habitat availability and species diversity in space and time. Due to the automated

implementation of the modeling framework used in this project, the methods developed and

used here can (with slight modifications) transferred to remote sensing data with other spatial

resolutions available in the near future. In summary, future research therefore should focus on

(1) the applicability of other remotely sensed parameters not considered here for predicting

species distributions, (2) differences in the performance of remote sensing-based SDMs for

species associated with different habitat types and (3) the potential of remote sensing data to

verify (and falsify) species occurrence records.

21

8. References

Araújo, M.B., Pearson, R.G., Thuiller, W. & Erhard, M. (2005). Validation of species-climate impact

models under climate change. Global Change Biology, 11, 1504-1513.

Bradley, B.A. & Fleishman, E. (2008). Can remote sensing of land cover improve species distribution

modelling?, Journal of Biogeography, 35 (7), 1158-1159.

Buermann, W., Saatchi, S., Smith, T.B., Zutta, B.R., Chavesi, J.A., Mila, B. & Graham, C.H. (2008).

Predicting species distributions across the Amazonian and Andean regions using remote sensing

data. Journal of Biogeography, 35 (7), 1160-1176.

Buisson, L., Thuiller, W., Casajus, N., Lek, S. & Grenouillet, G. (2010). Uncertainty in ensemble

forecasting of species distribution. Global Change Biology, 16, 1145-1157.

Colditz, R.R., Conrad, C., Wehrmann, T., Schmidt, M., & Dech, S. (2008). TiSeG: A flexible Software

Tool for Time-Series Generation of MODIS Data Utilizing the Quality Assessment Science Data

Set. IEEE Transactions on Geoscience and Remote Sensing, 46(10), 3296-3308.

Colditz, R.R., Cord, A., Conrad, C., Mora, F., Maeda, P. & Ressl, R. (2009). Analyzing Phenological

Characteristics of Mexico with MODIS Time Series Products. Multi-Temp, Fifth international

Workshop on the Analysis of Multitemporal Remote Sensing Images, July 28-30, 2009. Mystic,

Connecticut, USA.

Comisión Nacional para el Conocimiento y Uso de la Biodiversidad – CONABIO (1997). Provincias

biogeográficas de México. Escala 1:4000000. México.

Cord, A. & Rödder, D. (2011). Inclusion of habitat availability in species distribution models through

multi-temporal remote sensing data? Ecological Applications, 21, 3285-3298.

Dormann, C.F., Purschke, O. García Márquez, J. R., Lautenbach, S. & Schröder, B. (2008). Components

of uncertainty in species distribution analysis: A case study of the Great Grey Shrike Lanius

excubitor L. Ecology, 89, 3371-3386.

Dormann, C.F., Elith, J., Bacher, S., Buchmann, C., Carl, G., Carré, G., García Marquéz, J.R., Gruber,

B., Lafourcade, B., Leitao, P.J., Münkemüller, T., McClean, C., Osborne, P.E., Reineking, B.,

Schröder, B., Skidmore, A.K., Zurell, D. & Lautenbach, S. (2013) Collinearity: a review of

methods to deal with it and a simulation study to evaluate their performance. Ecography, 36,

27-46.

Eklundh, L. & Jönsson, P. (2009). Timesat 3.0, Software Manual, Lund University, Sweden.

Elith, J., Burgman M. A. & Regan, H. M. (2002). Mapping uncertainties and vague concepts in

predictions of species distributions. Ecological Modelling, 157, 313-329.

Farjon, A. & Filer, D. (2013). An Atlas of the World's Conifers: An Analysis of Their Distribution,

Biogeography, Diversity and Conservation Status. Brill Academic Pub, 512 pp.

Freeman, E. A. & Moisen, G. (2008). PresenceAbsence: An R Package for Presence-Absence

Model Analysis. Journal of Statistical Software, 23(11),1-31.

GADM (2012). GADM database of Global Administrative Areas, Version 2.0, www.gadm.org

Hartigan, J.A. & Wong, M.A. (1979). Algorithm AS 136: A K-Means Clustering Algorithm. Journal of

the Royal Statistical Society, Series C (Applied Statistics), 28, 100–108.

Hijmans, R.J. and Elith, J. (2015). Species distribution modeling with R, http://cran.r-

project.org/web/packages/dismo/vignettes/sdm.pdf

http://www.gadm.org/

http://cran.r-project.org/web/packages/dismo/vignettes/sdm.pdf

http://cran.r-project.org/web/packages/dismo/vignettes/sdm.pdf

22

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., & Jarvis, A. (2005). Very high resolution

interpolated climate surfaces for global land areas. International Journal of Climatology, 25,

1965-1978.

Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X. & Ferreira, L.G. (2002). Overview of the

radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of

Environment, 83, 195-213

Jarnevich, C.S. & Reynolds, L.V. (2011). Challenges of predicting the potential distribution of a slow-

spreading invader: a habitat suitability map for an invasive riparian tree. Biological Invasions,

13, 153-163.

Jönsson, P. & Eklundh, L. (2004). TIMESAT – a program for analyzing time-series of satellite sensor

data, Computers & Geosciences, 30, 833-845.

Leutner, B.F., Reineking, B., Müller, J., Bachmann, M., Beierkuhnlein, C., Dech, S. & Wegmann, M.

(2012). Modelling Forest α -Diversity and Floristic Composition On the Added Value of LiDAR

plus Hyperspectral Remote Sensing. Remote Sensing, 4, 2818-2845.

Livingston, F. (2005). Implementation of Breiman’s Random Forest Machine Learning Algorithm,

ECE591Q Machine Learning Journal Paper, https://datajobs.com/data-science-repo/Random-

Forest-%5BFrederick-Livingston%5D.pdf

Liu, C., Pam, M., Dawson, T.P. & Pearson, R.G. (2005). Selecting thresholds of occurrence in the

prediction of species distributions. Ecography, 28, 385-393.

Luoto, M., Virkkala, R. & Heikkinen, R.K. (2007). The role of land cover in bioclimatic models depends

on spatial resolution. Global Ecology and Biogeography, 16, 34-42.

Pearson, R. G., Dawson, T. P. & Liu, C. (2004). Modelling species distributions in Britain: a hierarchical

integration of climate and landcover data. Ecography, 27, 285-298.

Pebesma, E.J. & Bivand, R.S. (2005). Classes and methods for spatial data in R. R News 5 (2),

http://cran.r-project.org/doc/Rnews/.

Peterson, A.T., Papes, M. & Soberón, J. (2008). Rethinking receiver operating characteristic analysis

applications in ecological niche modeling. Ecological Modelling, 213, 63-72.

Phillips, S.J., Anderson, R.P., & Schapire, R.E. (2006). Maximum entropy modeling of species

geographic distributions. Ecological Modelling, 190, 231-259.

Phillips, S.J., Dudík, M., Elith, J., Graham, C.H., Lehmann, A., Leathwick, J., & Ferrier, S. (2009).

Sample selection bias and presence-only distribution models: Implications for background and

pseudo-absence data. Ecological Applications, 19, 181-197.

R Core Team (2013). R: A language and environment for statistical computing. R Foundation for

Statistical Computing, Vienna, Austria. http://www.R-project.org/.

R Core Team (2014). R: A language and environment for statistical computing. R Foundation for

Statistical Computing, Vienna, Austria. http://www.R-project.org/.

Robin, X., Turck, N., Hainard, A., Tiberti, N., Lisacek, F., Sanchez, J.-C. & Müller, M. (2011). pROC:

an open-source package for R and S+ to analyze and compare ROC curves. BMC Bioinformatics,

77.

Rzedowski, J. (1978). Vegetación de México. Editorial Limusa, México. I.

https://datajobs.com/data-science-repo/Random-Forest-%5BFrederick-Livingston%5D.pdf

https://datajobs.com/data-science-repo/Random-Forest-%5BFrederick-Livingston%5D.pdf

http://cran.r-project.org/doc/Rnews/

http://www.r-project.org/

http://www.r-project.org/

23

Rzedowski (1990). Vegetación Potencial, Atlas Nacional de México. Vol II. Escala 1:4 000 000.

Instituto de Geografía, UNAM. México

Saatchi, S., Buermann, W., ter Steege, H., Mori, S. & Smith, T.B. (2008). Modeling distribution of

Amazonian tree species and diversity using remote sensing measurements. Remote Sensing of

Environment, 112, 2000-2017.

Thuiller, W. Miguel B. Araújo, M.B. & Lavorel, S. (2004). Do we need land-cover data to model species

distributions in Europe?. Journal of Biogeography, 31, 353-361.

Thuiller, W., Georges, D. & Engler, R. (2014). biomod2: Ensemble platform for species distribution

modeling. R package version 3.1-48. http://CRAN.R-project.org/package=biomod2.

Tuanmu, M.-N., Viña, A., Bearer, S., Xu, W., Ouyang, Z., Zhang, H. & Liu, J. (2010). Mapping

understory vegetation using phenological characteristics derived from remotely sensed data,

Remote Sensing of Environment, 114, 1833-1844.

Venables, W. N. & Ripley, B. D. (2002). Modern Applied Statistics with S. Fourth Edition. Springer,

New York. ISBN 0-387-95457-0.

9. Appendix

1. Documentation of delivered materials (Documentación del material

entregado)

See the following pages

http://cran.r-project.org/package=biomod2

24

A1. Species considered in this project. Overview of species considered and modeled in the project.

Family Names of included

species (NOM-059)

Accepted names of included species

(project JM078)

Species

with

models

Species

without

models

Short name

(maps and

metadata)

Unique

records

in

database

(covered

by

climate

data)

Model

algorithm

(s) used

for final

climate-

based

models

Unique

records in

database

(covered

by remote

sensing

data)

Model

algorith

m(s)

used for

final

remote

sensing-

based

models

Pinaceae Abies concolor Abies concolor x ABICON 577 Maxent,

GLM, RF 10

Pinaceae Abies flinckii Abies flinckii x 6 16

Pinaceae Abies guatemalensis Abies guatemalensis x 15 14

Pinaceae Abies hickelii Abies hickelii x 16 15

Pinaceae Abies vejari Abies vejarii x ABIVEJ 27 Maxent 24 Maxent

Pinaceae Abies vejari mexicana Abies vejarii subsp. mexicana x 0 0

Orchidaceae Acianthera eximia Acianthera eximia (L.O. Williams)

Solano x 8 9

Orchidaceae Acianthera unguicallosa Acianthera unguicallosa (Ames & C.

Schweinf.) Solano x 2 2

Orchidaceae Acianthera violacea Acianthera violacea (A. Rich. &

Galeotti) Pridgeon & M.W. Chase x 0 0

Orchidaceae Acineta barkeri Acineta barkeri (Bateman) Lindl. x 6 5

Orchidaceae Anathallis abbreviata Anathallis abbreviata (Schltr.)

Pridgeon & M.W. Chase x 0 0

Orchidaceae Anathallis oblanceolata Anathallis oblanceolata (L.O.

Williams) Solano & Soto Arenas x 1 1

Orchidaceae Aspidogyne stictophylla Aspidogyne stictophylla (Schltr.) Garay x 12 10

Orchidaceae Barbosella prorepens Barbosella prorepens (Rchb. f.) Schltr. x 2 2

Orchidaceae Barkeria dorotheae Barkeria dorotheae Halb. x 4 4

Orchidaceae Barkeria melanocaulon Barkeria melanocaulon A. Rich. &

Galeotti x 11 13

Orchidaceae Barkeria scandens Barkeria scandens (Lex.) Dressler &

Halb. x 19 19

25

Orchidaceae Barkeria shoemakeri Barkeria shoemakeri Halb. x 0 0

Orchidaceae Barkeria skinneri Barkeria skinneri (Bateman ex Lindl.)

Lindl. ex Paxton x 4 7

Orchidaceae Barkeria strophinx Barkeria strophinx (Rchb. f.) Halb. x 2 2

Orchidaceae Barkeria whartoniana Barkeria whartoniana (C. Schweinf.)

Soto Arenas x 3 7

Orchidaceae Bletia urbana Bletia urbana Dressler x 3 8

Cupressaceae Calocedrus decurrens Calocedrus decurrens x CALDEC 109

Maxent,

GLM,

RF

9

Orchidaceae Caularthron

bilamellatum

Caularthron bilamellatum (Rchb. f.)

R.E. Schultes x 0 0

Orchidaceae Chysis bractescens Chysis bractescens Lindl. x 12 18

Orchidaceae Chysis limminghei Chysis limminghei Linden & Rchb. f. x 3 2

Orchidaceae Clowesia glaucoglossa Clowesia glaucoglossa (Rchb. f.)

Dodson x 2 1

Orchidaceae Clowesia rosea Clowesia rosea Lindl. x 7 7

Orchidaceae Cochleanthes

flabelliformis

Cochleanthes flabeliformis (Sw.) R.E.

Schultes & Garay x 0 0

Orchidaceae Coelia densiflora Coelia densiflora Rolfe x 5 5

Orchidaceae Corallorhiza macrantha Corallorhiza macrantha Schltr., x 9 9

Orchidaceae Cryptarrhena lunata Cryptarrhena lunata R. Br. x 0 0

Orchidaceae Cuitlauzina candida Cuitlauzina candida (Lindl.) Dressler

& N.H. Williams, 2003 x 3 3

Orchidaceae Cuitlauzina

convallarioides

Cuitlauzina convallarioides (Schltr.)

Dressler & N.H. Williams x 0 0

Orchidaceae Cuitlauzina pendula Cuitlauzina pendula Lex. x 10 11

Cupressaceae Cupressus arizonica

montana Callitropsis montana x 0 0

Cupressaceae Cupressus forbesii Callitropsis forbesii x CUPFOR 36 Maxent 0

Cupressaceae Cupressus guadalupensis Callitropsis guadalupensis x 2 2

Cupressaceae Cupressus lusitanica Callitropsis lusitanica x CUPLUS 111

Maxent,

GLM,

RF

106

Maxent,

GLM,

RF

Orchidaceae Cycnoches ventricosum Cycnoches ventricosum Bateman x CYCVEN 23 Maxent 23 Maxent

26

Orchidaceae Cypripedium

dickinsonianum Cypripedium dickinsonianum Hágsater x 0 0

Orchidaceae Cypripedium irapeanum Cypripedium irapeanum Lex. x 0 0

Orchidaceae Cyrtochiloides

ochmatochila

Cyrtochiloides ochmatochila (Rchb.f.)

N.H. Williams & M.W. Chase x 3 3

Orchidaceae Dignathe pygmaeus Dignathe pygmaeus Lindl. x 0 0

Orchidaceae Dracula pusilla Dracula pusilla (Rolfe) Luer x 3 2

Orchidaceae Dryadella guatemalensis Dryadella guatemalensis (Schltr.) Luer x 5 6

Orchidaceae Elleanthus

hymenophorus

Elleanthus hymenophorus (Rchb. f.)

Rchb. f. x 0 0

Orchidaceae Encyclia adenocaula Encyclia adenocaula (Lex.) Schltr. x ENCADE 46

Maxent,

GLM,

RF

45

Maxent,

GLM,

RF

Orchidaceae Encyclia atrorubens Encyclia atrorubens (Rolfe) Schltr. x 0 0

Orchidaceae Encyclia distantiflora Oestlundia distantiflora (A. Rich. &

Galeotti) Dressler & Pollard x 0 0

Orchidaceae Encyclia kienastii Encyclia kienastii (Rchb. f.) Dressler &

G.E. Pollard x 5 7

Orchidaceae Encyclia lorata Encyclia lorata Dressler & G.E.

Pollard x 1 1

Orchidaceae Encyclia pollardiana Encyclia pollardiana (Withner)

Dressler & G.E. Pollard x 7 7

Orchidaceae Encyclia tuerckheimii Encyclia tuerckheimii Schltr. x 2 5

Orchidaceae Epidendrum

alabastrialatum

Epidendrum alabastrialatum

G.E.Pollard ex Hágsater x 7 5

Orchidaceae Epidendrum alticola Epidendrum alticola Ames & Correll x 4 5

Orchidaceae Epidendrum cerinum Epidendrum cerinum Schltr. x 7 8

Orchidaceae Epidendrum chloe Epidendrum chloe Rchb. f. x 5 6


cnemidophorum Epidendrum cnemidophorum Lindl. x 15 18

Orchidaceae Epidendrum coronatum Epidendrum coronatum Ruiz & Pav. x 1 1

Orchidaceae Epidendrum cystosum Epidendrum cystosum Ames x 3 3


dorsocarinatum Epidendrum dorsocarinatum Hágsater x 2 2

Orchidaceae Epidendrum dressleri Epidendrum dressleri Hágsater x 0 0

27


incomptoides

Epidendrum incomptoides Ames, F.T.

Hubb. & C. Schweinf. x 1 1

Orchidaceae Epidendrum isthmii Epidendrum isthmi Schltr. x 0 0

Orchidaceae Epidendrum pansamalae Epidendrum pansamalae Schltr. x 4 4

Orchidaceae Epidendrum skutchii Epidendrum skutchii Ames, F.T. Hubb.

& C. Schweinf. x 5 5


smaragdinum Epidendrum smaragdinum Lindl. x 5 5

Orchidaceae Epidendrum sobralioides Epidendrum sobralioides Ames &

Correll x 0 0

Orchidaceae Erycina crista-galli Erycina crista-galli (Rchb. f.) N.H.

Williams & M.W. Chase x 12 11

Orchidaceae Erycina pumilio Erycina pumilio (Rchb. f.) N.H.

Williams & M.W. Chase x 0 0

Orchidaceae Euchile citrina Prosthechea citrina (Lex.) W.E.

Higgins x 0 0

Orchidaceae Euchile mariae Prosthechea mariae (Ames) Hoehne x 0 0

Orchidaceae Eurystyles borealis Eurystyles borealis A.H. Heller x 0 0

Orchidaceae Galeandra batemanii Galeandra batemanii Rolfe x 3 5

Orchidaceae Galeottia grandiflora Galeottia grandiflora A.Rich. x 0 0

Orchidaceae Galeottiella sarcoglossa Galeottiella sarcoglossa (A. Rich. &

Galeotti) Schltr. x 11 11

Orchidaceae Gongora tridentata Gongora tridentata Whitten x 0 0

Orchidaceae Govenia tequilana Govenia tequilana Dressler &

Hágsater x 0 0

Orchidaceae Guarianthe skinneri Guarianthe skinneri (Bateman)

Dressler & W.E. Higgins x 5 6

Orchidaceae Habenaria umbratilis Habenaria umbratilis Ames & L.O.

Williams x 0 1

Orchidaceae Hagsatera

brachycolumna

Hagsatera brachycolumna (L.O.

Williams) R. González x 8 10

Orchidaceae Ionopsis satyrioides Ionopsis satyrioides (Sw.) Rchb. f. x 0 0

Orchidaceae Jacquiniella gigantea Jacquiniella gigantea Dressler, Salazar

& García-Cruz x 2 2

Cupressaceae Juniperus californica Juniperus californica x JUNCAL 37 Maxent 16

Cupressaceae Juniperus sabinoides

monticola Juniperus monticola x JUNSAB 23 Maxent 24 Maxent

28

Orchidaceae Kefersteinia lactea Kefersteinia tinschertiana Pupulin x 0 0

Orchidaceae Kraenzlinella hintonii Kraenzlinella hintonii (L.O. Williams)

Solano x 2 2

Orchidaceae Lacaena bicolor Lacaena bicolor Lindl. x 0 0

Orchidaceae Laelia anceps dawsonii Laelia anceps Lindl. subsp. dawsonii

(J. Anderson) Rolfe x 10 10

Orchidaceae Laelia gouldiana Laelia gouldiana Rchb. f. x 7 7

Orchidaceae Laelia speciosa Laelia speciosa (Kunth) Schltr. x LAESPE 43

Maxent,

GLM,

RF

43

Maxent,

GLM,

RF

Orchidaceae Laelia superbiens Laelia superbiens Lindl. x 13 12

Orchidaceae Lepanthes ancylopetala Lepanthes ancylopetala Dreesler x 1 1

Orchidaceae Lepanthes guatemalensis Lepanthes guatemalensis Schltr. x 2 2

Orchidaceae Lepanthes parvula Lepanthes parvula Dressler x 2 2

Orchidaceae Lepanthopsis floripecten Lepanthopsis floripecten (Rchb. f)

Ames x 2 3

Orchidaceae Ligeophila clavigera Ligeophila clavigera (Rchb. f.) Garay x 11 12

Orchidaceae Lycaste lasioglossa Lycaste lasioglossa Rchb. f. x 1 1

Orchidaceae Lycaste skinneri Lycaste skinneri (Bateman ex Lindl.)

Lindl. x 11 11

Orchidaceae Lyroglossa pubicaulis Lyroglossa pubicaulis (L.O.Williams)

Garay x 0 0

Orchidaceae Macradenia brassavolae Macradenia brassavolae Rchb. f. x 0 0

Orchidaceae Malaxis greenwoodiana Malaxis greenwoodiana Salazar &

Soto Arenas x 4 4

Orchidaceae Malaxis hagsateri Malaxis hagsateri Salazar x 5 4

Orchidaceae Malaxis pandurata Malaxis pandurata (Schltr.) Ames x 0 0

Orchidaceae Maxillaria alba Maxillaria alba (Hook.) Lindl. x 4 3

Orchidaceae Maxillaria nasuta Maxillaria nasuta Rchb. f. x 3 3

Orchidaceae Maxillaria oestlundiana Maxillaria oestlundiana L.O.Williams x 0 0

Orchidaceae Maxillaria tonsoniae Maxillaria tonsoniae Soto-Arenas x 11 11

Orchidaceae Mexipedium

xerophyticum

Mexipedium xerophyticum (Soto

Arenas, Salazar & Hágsater) V.A.

Albert & M.W. Chase

x 0 0

29

Orchidaceae Mormodes maculata

unicolor

Mormodes maculata (Klotzch) L.O.

Williams var. unicolor (Klotzsch) L.O.

Williams

x MORMAC 19 Maxent 19 Maxent

Orchidaceae Mormodes

porphyrophlebia Mormodes porphyrophlebia Salazar x 0 0

Orchidaceae Mormodes

sanguineoclaustra Mormodes sanguineoclaustra Fowl. x 3 3

Orchidaceae Mormodes sotoana Mormodes sotoana Salazar x 1 1

Orchidaceae Mormodes uncia Mormodes uncia Rchb. f. x 0 0

Orchidaceae Nemaconia dressleriana Nemaconia dressleriana Soto Arenas x 0 0

Orchidaceae Nemaconia pellita Nemaconia pellita (Rchb.f.) van den

Berg, Salazar & Soto Arenas x 0 0

Orchidaceae Oestlundia distantiflora Oestlundia distantiflora (A. Rich. &

Galeotti) Dressler & Pollard x 7 5

Orchidaceae Oncidium endocharis Oncidium endocharis Rchb.f. x 5 5

Orchidaceae Oncidium ensatum Oncidium ensatum Lindl. x 7 7

Orchidaceae Oncidium exauriculatum Oncidium exauriculatum (Hamer &

Garay) Jiménez x 1 1

Orchidaceae Oncidium incurvum Oncidium incurvum Barker ex Lindl. x ONCINC 35 Maxent 39 Maxent

Orchidaceae Oncidium leucochilum Oncidium leucochilum Barker ex Lindl. x ONCLEU 27 Maxent 30 Maxent

Orchidaceae Oncidium

ochmatochilum Oncidium ochmatochilum Rchb.f. x 0 0

Orchidaceae Oncidium pollardii Oncidium iricolor Rchb. f. x 5 6

Orchidaceae Oncidium stelligerum Oncidium stelligerum Rchb. f. x 9 9

Orchidaceae Oncidium suttonii Oncidium suttonii Batem ex Lindley x 0 0

Orchidaceae Oncidium tigrinum Oncidium tigrinum Lex. x ONCTIG 23 Maxent 22 Maxent

Orchidaceae Oncidium unguiculatum Oncidium unguiculatum Lindl. x ONCUNG 24 Maxent 26 Maxent

Orchidaceae Oncidium

wentworthianum

Oncidium wentworthianum Bateman ex

Lindl. x 3 3

Orchidaceae Osmoglossum

convallarioides

Cuitlauzina convallarioides (Schltr.)


Orchidaceae Pachyphyllum

mexicanum

Pachyphyllum mexicanum Dressler &

Hágsater x 6 6

Orchidaceae Palumbina candida Cuitlauzina candida (Lindl.) Dressler

& N.H. Williams x 0 0

30

Orchidaceae Papperitzia leiboldii Leochilus leiboldii Rchb. f. x 4 4

Orchidaceae Pelexia congesta Pelexia congesta Ames & Schweinf. x 2 1

Orchidaceae Phragmipedium

exstaminodium

Phragmipedium humboldtii J.T.

Atwood & Dressler subsp.

exstaminodium (Castaño, Hágsater &

E.Aguirre) J.T. Atwood & Dressler ex

J.M.H. Shaw

x 0 0

Orchidaceae Physogyne gonzalezii Physogyne gonzalezii (L.O. Williams)

Garay x 1 1

Pinaceae Picea chihuahuana Picea chihuahuana x 12 15

Pinaceae Picea engelmannii

mexicana Picea mexicana x 6 7

Pinaceae Picea martinezii Picea martinezii x 0 0

Pinaceae Pinus attenuata Pinus attenuata x PINATT 103

Maxent,

GLM,

RF

7

Pinaceae Pinus caribaea

hondurensis Pinus caribaea var. hondurensis x PINCAR 72

Maxent,

GLM,

RF

6

Pinaceae Pinus contorta

murrayana Pinus contorta subsp. murrayana x PINCON 52

Maxent,

GLM,

RF

10

Pinaceae Pinus coulteri Pinus coulteri x PINCOU 98

Maxent,

GLM,

RF

9

Pinaceae Pinus culminicola Pinus culminicola x PINCUL 31 Maxent 24 Maxent

Pinaceae Pinus jaliscana Pinus jaliscana x 5 4

Pinaceae Pinus jeffreyi Pinus jeffreyi x PINJEF 98

Maxent,

GLM,

RF

18

Pinaceae Pinus johannis Pinus johannis x PINJOH 23 Maxent 23 Maxent

Pinaceae Pinus lagunae Pinus lagunae x 0 0

Pinaceae Pinus martinezii Pinus douglasiana x 0 0

Pinaceae Pinus maximartinezii Pinus maximartinezii x 16 18

Pinaceae Pinus monophylla Pinus californiarum x PINCAL 39 Maxent 20 Maxent

31

Pinaceae Pinus muricata Pinus muricata x PINMUR 86

Maxent,

GLM,

RF

9

Pinaceae Pinus nelsonii Pinus nelsonii x PINNEL 27 Maxent 27 Maxent

Pinaceae Pinus pinceana Pinus pinceana x PINPIN 59

Maxent,

GLM,

RF

59

Maxent,

GLM,

RF

Pinaceae Pinus quadrifolia Pinus quadrifolia x PINQUA 47

Maxent,

GLM,

RF

40

Maxent,

GLM,

RF

Pinaceae Pinus remota Pinus remota x 16 13

Pinaceae Pinus rzedowskii Pinus rzedowskii x 8 8

Pinaceae Pinus strobiformis Pinus strobiformis x PINSTR 59

Maxent,

GLM,

RF

49

Maxent,

GLM,

RF

Pinaceae Pinus strobus chiapensis Pinus chiapensis x PINSTRC

HI 41

Maxent,

GLM,

RF

40

Maxent,

GLM,

RF

Orchidaceae Platystele caudatisepala Platystele caudatisepala (C. Schwein

f.) Garay x 1 1

Orchidaceae Platystele

jungermannioides

Platystele jungermannioides (Schltr.)

Garay x 0 0

Orchidaceae Platystele repens Platystele repens (Ames) Garay x 0 0

Orchidaceae Platythelys venustula Platythelys venustula (Ames) Garay x 0 0

Orchidaceae Pleurothallis hintonii Kraenzlinella hintonii (L.O. Williams)

Solano x 0 0

Orchidaceae Pleurothallis nelsonii Pleurothallis nelsonii Ames x 3 3

Orchidaceae Pleurothallis

saccatilabia Pleurothallis saccatilabia C. Schweinf. x 0 0

Orchidaceae Pleurothallopsis

ujarensis

Pleurothallopsis ujarensis (Rchb.f.)


Podocarpaceae Podocarpus matudae Podocarpus matudae x PODMAT 52

Maxent,

GLM,

RF

52

Maxent,

GLM,

RF

Orchidaceae Ponthieva brittoniae Ponthieva brittoniae Ames & C.

Schwinf. x 3 4

Orchidaceae Prosthechea abbreviata Prosthechea abbreviata (Schltl.) W.E.

Higgins x 0 0

32

Orchidaceae Prosthechea citrina Prosthechea citrina (Lex.) W.E.

Higgins x PROCIT 41

Maxent,

GLM,

RF

36 Maxent

Orchidaceae Prosthechea mariae Prosthechea mariae (Ames) W.E.

Higgins x 13 11

Orchidaceae Prosthechea neurosa Prosthechea neurosa (Ames) W.E.

Higgins x 10 10

Orchidaceae Prosthechea vagans Prosthechea vagans (Ames) W.E.

Higgins x 0 0

Orchidaceae Prosthechea vitellina Prosthechea vitellina (Lindl.) W.E.

Higgins x PROVIT 86

Maxent,

GLM,

RF

83

Maxent,

GLM,

RF

Orchidaceae Pseudocranichis

thysanochila

Galeoglossum thysanochilum (B.L.

Rob. & Greenm.) Salazar x 0 4

Orchidaceae Pseudogoodyera

pseudogoodyeroides

Pseudogoodyera pseudogoodyeroides

(Rchb. f.) Schltr. x 4 0

Pinaceae Pseudotsuga menziesii

glauca Pseudotsuga menziesii var. glauca x PSEMEN 611

Maxent,

GLM,

RF

87

Maxent,

GLM,

RF

Orchidaceae Restrepia lankesteri Restrepia trichoglossa F.Lehm. ex

Sander x 0 0

Orchidaceae Rhynchostele beloglossa Rhynchostele beloglossa (Rchb. f.)


Orchidaceae Rhynchostele cervantesii Rhynchostele cervantesii (Lex.) Soto

Arenas & Salazar x RHYCER 95

Maxent,

GLM,

RF

94

Maxent,

GLM,

RF

Orchidaceae Rhynchostele cordata Rhynchostele cordata (Lindl.) Soto

Arenas & Salazar x RHYCOR 46

Maxent,

GLM,

RF

42

Maxent,

GLM,

RF

Orchidaceae Rhynchostele

ehrenbergii

Rhynchostele ehrenbergii (Link,

Klotzsch & Otto) Soto Arenas &

Salazar

x RHYEHR 24 Maxent 21 Maxent

Orchidaceae Rhynchostele galeottiana Rhynchostele galeottiana (A. Rich.)

Soto Arenas & Salazar x 0 0

Orchidaceae Rhynchostele

londesboroughiana

Rhynchostele londesboroughiana

(Rchb.f.) Halbinger x 5 5

Orchidaceae Rhynchostele madrensis Rhynchostele madrensis (Rchb.f.) Soto

Arenas & Salazar x 11 9

33

Orchidaceae Rhynchostele majalis

Rhynchostele majalis (Rchb.f.)

Halbinger y (Rchb.f.) Soto Arenas &

Salazar

x 4 4

Orchidaceae Rhynchostele pygmaea Rhynchostele pygmaea (Lindl.) Rchb. f. x 8 7

Orchidaceae Rhynchostele rossii Rhynchostele rossii (Lindl.) Soto

Arenas & Salazar x RHYROS 43

Maxent,

GLM,

RF

41

Maxent,

GLM,

RF

Orchidaceae Rhynchostele uroskinneri Rhynchostele uroskinneri (Lindl.) Soto

Arenas & Salazar x 0 0

Orchidaceae Rodriguezia dressleriana Rodriguezia dressleriana R. González x 2 2

Orchidaceae Rossioglossum grande Rossioglossum grande (Lindl.) Garay

& G.C.Kenn. x 5 6

Orchidaceae Rossioglossum insleayi Rossioglossum insleayi (Barker ex

Lindl.) Garay & G.C. Kenn. x 0 0

Orchidaceae Rossioglossum splendens Rossioglossum splendens (Rchb.f.)

Garay & G.C. Kenn. x 0 0

Orchidaceae Rossioglossum

williamsianum

Rossioglossum williamsianum (Rchb.

f.) Garay & G.C. Kenn. x 0 0

Orchidaceae Sarcoglottis cerina Sarcoglottis cerina (Lindl.) P.N. Don x 3 4

Orchidaceae Scelochilus tuerckheimii Comparettia tuerckheimii (Schltr.)

M.W. Chase & N.H. Williams x 0 0

Orchidaceae Schiedeella nagelii Schiedeella nagelii (L.O. Williams)

Garay x 6 6

Orchidaceae Sigmatostalix

guatemalensis

Oncidium poikilostalix (Kraenzl.)


Orchidaceae Sigmatostalix mexicana Oncidium mexicanum (L.O. Williams)


Orchidaceae Sobralia crispissima Sobralia crispissima Dressler x 1 1

Orchidaceae Sobralia mucronata Sobralia mucronata Ames x 4 4

Orchidaceae Specklinia digitale Specklinia digitale (Luer) Pridgeon &

M.W. Chase x 19 16

Orchidaceae Specklinia endotrachys Specklinia endotrachys (Rchb. f.)


Orchidaceae Specklinia fimbriata Specklinia fimbriata C. Schweinf. &

(Ames & C. Schweinf.) Solano x 4 4

Orchidaceae Specklinia glandulosa Specklinia glandulosa (Ames) Pridgeon

& M.W. Chase x 0 0

34

Orchidaceae Specklinia lateritia Specklinia lateritia (Rchb.f.) Pridgeon

& M.W. Chase x 0 0

Orchidaceae Spiranthes torta Spiranthes torta (Thunb.) Garay &

H.R. Sweet x 1 1

Orchidaceae Stanhopea ecornuta Stanhopea ecornuta Lem. x 2 1

Orchidaceae Stanhopea oculata Stanhopea oculata (G.Lodd.) Lindl. x STAOCU 22 Maxent 20 Maxent

Orchidaceae Stanhopea tigrina Stanhopea tigrina Bateman x STATIG 24 Maxent 24 Maxent

Orchidaceae Stelis chihobensis Stelis chihobensis Ames x 7 7

Orchidaceae Stelis cobanensis Stelis cobanensis (Schltr.) Pridgeon &

M.W. Chase x 18 18

Orchidaceae Stelis deregularis Stelis deregularis Barb. Rodr. x 3 2

Orchidaceae Stelis nigriflora Stelis nigriflora (L.O. Williams)


Orchidaceae Stellilabium standleyi Telipogon standleyi Ames x 1 1

Taxaceae Taxus globosa Taxus globosa x TAXGLO 51

Maxent,

GLM,

RF

51

Maxent,

GLM,

RF

Orchidaceae Teuscheria pickiana Teuscheria pickiana (Schltr.) Garay x 0 0

Orchidaceae Trichocentrum

flavovirens

Trichocentrum flavovirens L.O.

Williams x 6 6

Orchidaceae Trichocentrum hoegei Trichocentrum hoegei Rchb. f. x 9 8

Orchidaceae Trichocentrum

stramineum

Trichocentrum stramineum (Bateman

ex Lindl.) M.W. Chase & N.H. Williams x 14 16

Orchidaceae Trichopilia galeottiana Trichopilia galeottiana A.Rich. &

Galeotti x 3 2

Orchidaceae Trichopilia subulata Trichopilia subulata (Sw.) Rchb.f. x 0 0

Orchidaceae Trichosalpinx

cedralensis Trichosalpinx cedralensis (Ames) Luer x 0 0

Orchidaceae Vanilla planifolia Vanilla planifolia Jacks. x VANPLA 32 Maxent 32 Maxent

Orchidaceae Warrea costaricensis Warrea costaricensis Schltr. x 1 1

35

2. Quality of information (Calidad de información)

The following Appendices (A2, A3, A4 and A5) are submitted as separate files together with this

report.

A2. Climatic profiles (table). Summary of climatic conditions at the presence localities used in

the species distribution models (mean and standard deviation).

A3. Climatic profiles (boxplots). Summary of climatic conditions at the presence localities used

in the species distribution models (mean and standard deviation).

A4 and A5. Profiles of land surface phenology (tables). Species-specific phenological profiles

obtained from remotely sensed time series of the Enhanced Vegetation Index (a) for all occurrence

localities together, (b) stratified by biogeographic provinces and (c) stratified by vegetation types.

A5. Profiles of land surface phenology (figures). Species-specific phenological profiles obtained

from remotely sensed time series of the Enhanced Vegetation Index for all occurrence localities

together and stratified by biogeographic provinces and vegetation types.

36

3. Other information

A6. Overview of how often each climate variable was selected for the species distribution

models (as a result of the collinearity and variable importance analysis). 0.293 means that the

variable was selected in 29.3% of the models.

Variable Mean Standard deviation

BIO1 0.000 0.000

BIO2 0.293 0.461

BIO3 0.073 0.264

BIO4 0.439 0.502

BIO5 0.610 0.494

BIO6 0.024 0.156

BIO7 0.049 0.218

BIO8 0.317 0.471

BIO9 0.146 0.358

BIO10 0.098 0.300

BIO11 0.024 0.156

BIO12 0.000 0.000

BIO13 0.073 0.264

BIO14 0.780 0.419

BIO15 1.000 0.000

BIO16 0.122 0.331

BIO17 0.220 0.419

BIO18 0.415 0.499

BIO19 1.000 0.000

37

A7. Overview of how often each remote sensing variable was selected for the species

distribution models (as a result of the collinearity and variable importance analysis). 0.677 means

that the variable was selected in 67.7% of the models.

Variable Mean Standard deviation

EVI_CoV 0.677 0.475

EVI_integral 0.000 0.000

EVI_maximum 0.065 0.250

EVI_mean 0.000 0.000

EVI_median 0.129 0.341

EVI_minimum 0.129 0.341

EVI_range 0.000 0.000

EVI_rate_greenup 0.194 0.402

EVI_rate_senescence 0.000 0.000

EVI_standard_deviation 0.129 0.341

EVI_value_EOS 0.032 0.180

EVI_value_SOS 0.032 0.180

LST_CoV 0.129 0.341

LST_maximum 0.194 0.402

LST_mean 0.258 0.445

LST_median 0.226 0.425

LST_minimum 0.903 0.301

LST_range 0.065 0.250

LST_standard_deviation 0.032 0.180

38

Additional material delivered. The following additional material was delivered upon completion

of the project:

- Database of the georeferenced localities of the 227 species considered in the project

(DarwinCore format, compiled in the context of project JM078)

- Detailed description of the variables used for distribution modeling (tabular format)

- Species-specific overview of variables used in the species distribution models (tabular format)

- Variable importance: Contribution of environmental variables to explaining species

distribution patterns for both climate and remote sensing data (tabular format)

- Model evaluation (based on AUC, partial AUC and model deviance) (tabular format)

- R-Code for species distribution modeling including remote sensing data and model

uncertainty analysis

- Digital maps of distributions of the species modeled in the project:

(a) Climatic ranges (for distribution range from southern USA to Central America):

1. Continuous probability of occurrence

2. Binary distribution map at minimum training presence threshold

3. Binary distribution map at 10 percentile training presence threshold

4. Binary distribution map at maximum training sensitivity plus specificity threshold

(b) Habitat availability within bioclimatic range derived from remote sensing time series

(only for Mexico)

1. Continuous probability of occurrence

2. Binary distribution map at minimum training presence threshold

3. Binary distribution map at 10 percentile training presence threshold

4. Binary distribution map at maximum training sensitivity plus specificity threshold

(c) Species-specific maps of habitat loss computed from climatic distribution ranges and

remotely sensed land surface properties

(d) Cumulative habitat loss across all study species

(e) Species-specific uncertainty maps based on the difference between the 97.5-percentile

and the 2.5-percentile of values derived from the continuous prediction maps (based on

199 model repetition runs)

Date post:	28-Feb-2019
Category:	Documents
Upload:	leliem
View:	214 times
Download:	0 times

Informe final* del Proyecto JM033 spatial measures of model … · Information on species...

Documents